How to make a song from text with Suno's generator

Type this into Suno's Style box: melancholic indie folk, fingerpicked acoustic guitar, brushed drums, warm analog tape, breathy female vocal, slow tempo. Paste eight lines of lyrics underneath. A minute later you have a finished track: a singer, a chord progression, a mix you could drop onto a playlist. That is the entire mechanic. Whether the song lands or gets skipped comes down to how you fill those two boxes.

Simple mode guesses, Custom mode obeys

Suno gives you two front doors. Simple mode is one box: you write "a sad folk song about leaving the town you grew up in", and Suno invents the structure, the words, and the sound all at once. It is fast, and it is a slot machine. Custom mode splits the job into a Style field and a Lyrics field, and that split is the whole reason to reach for it. In Custom you say how the song should sound in one field and set what it sings in the other, so the two stop fighting each other.

Rule of thumb: use Simple when you want to be surprised. Use Custom when you already hear the song in your head and need Suno to stop improvising. If you care about the result, it is Custom.

The Style field is a comma list, not a sentence

Style wants English descriptors separated by commas, not prose. A tidy list of tags reads far better to Suno than a paragraph. A strong Style line usually covers five slots:

Two things people get wrong here. First, they type a BPM number. Do not: describe tempo in words, slow tempo or driving uptempo, because a hard number tends to fight the groove and Suno mostly ignores it anyway. Second, they lock a musical key. Skip it. Naming a key rarely helps and often corners the model into a worse melody.

A worked example, from sketch to prompt

Say the idea is a note in your phone: "a late-night drive song, someone leaving a city they loved, hopeful but tired." That is a mood, not a prompt yet. Here is the translation into the two fields.

Style field:

nostalgic synthpop, midtempo, pulsing analog synth bass, gated reverb drums, shimmering pads, warm male vocal, 1980s production, bittersweet

Now the lyrics. You do not need finished verse to start, you need a skeleton with section tags so Suno knows the shape of the song. Anything in square brackets is a stage direction and never gets sung. Everything else is sung exactly as written.

[Intro]

[Verse]
Dashboard glow on an empty street
The city lights are losing heat

[Chorus]
I'm driving out, I'm driving slow
Down every road I'll ever know

[Verse]
Boxes in the back seat rattle
Every mile another battle

[Bridge]

[Outro]

Leaving [Intro], [Bridge], and [Outro] empty tells Suno to play instrumental sections there. That is how you get a real intro and a proper ending instead of a song that starts cold and cuts off mid-word.

Description only, or description plus lyrics

Those are your two ways to run it, and they buy different things.

Leave Lyrics blank and Suno still sings, it just makes up the words, and they are usually filler. For an instrumental, put instrumental in the Style field or flip the instrumental toggle, and it skips vocals entirely. For anything where the meaning matters, write at least a skeleton.

Generate, then judge the two takes

Every generation hands back two versions of the same prompt. They differ because randomness is baked into the process: the same Style and Lyrics seed two different melodies and two different vocal performances. That is a feature. Play both, keep the one with the stronger chorus, bin the other.

Here is the discipline that saves credits. Each run costs credits, and the free tier gives you a fixed allowance per day. Do not reroll ten times hoping for magic. If both takes are 80 percent right, stop spinning the wheel and switch to editing. If both are wrong the same way, the prompt is the problem, not the luck. Change one thing in the Style line, a single descriptor, and run again so you can see what moved.

Extend and refine instead of rerolling forever

Once a take is close, you stop generating from scratch. Suno's edit tools are where a rough draft turns into a track:

• Extend continues the song past where it stopped, so a two-minute idea grows into a full arrangement.
• Replace / Edit swaps a weak verse or a fumbled line without touching the rest.
• Cover keeps your melody and lyrics but re-records them in a new style.
• Personas capture the voice and vibe of a take you love, so the same singer can carry your other songs. That is how you build one recognizable artist instead of a pile of unrelated one-offs.

A blank Style field is the slowest way to start. If you would rather begin from a prompt that already sounds like a real genre, sunomarket keeps a catalog of tested Suno style prompts you can lift and adjust instead of guessing from an empty box.

FAQ

How long are Suno songs? A single generation usually lands in the low minutes, and newer model versions stretch that further. Want a full three or four minute arrangement? Generate the first section and use Extend to continue it. One click will not hand you a finished radio edit.

Can Suno make instrumentals with no vocals? Yes. Turn on the instrumental option or drop instrumental into the Style field, leave Lyrics empty, and you get music with no singing. Handy for background beds, loops, and demos where you will add your own voice later.

Why are the two takes so different? Same prompt, different roll. Suno adds randomness on every generation, so two versions from one Style line can split on melody, on how the tempo feels, and on the vocal. Treat them as two auditions, not two copies of one thing.

Can I sell what I make? Depends on your plan. The free tier is personal and non-commercial; paid plans generally add commercial rights to what you generate. Who owns purely AI-made music is still legally unsettled, so read Suno's current terms before you build a business on a track, and for real money, talk to a professional.